Security & privacy
Your records are safe with us, and they stay yours
You're trusting us with your organization's history and, often, sensitive details about the people you serve. Here's exactly how we use your data, how we protect it, and the controls your team has.
AI
How we use AI
Grant Autopilot uses Claude, a large language model made by Anthropic, through Anthropic's commercial API. Under that API, the data we send is not used to train Anthropic's models.
What the AI does
- Suggests facts from your documents and interview answers, each with the quote it came from, for your team to confirm.
- Asks interview questions and follow-ups.
- Reads a funder's requirements and estimates how well an opportunity fits.
- Drafts application sections from your facts.
- Checks the draft in a separate step, claim by claim, against the facts it cites.
What it never does
- Learn from your data. Your documents, interviews and drafts are never used to train AI models, ours or anyone else's.
- See facts you've marked restricted. Those are never sent to the AI when writing applications.
- Decide on its own to send an application. A person approves, unless your organization turns on autopilot for small requests and sets its limits.
- Share what it learns about you with another organization, or use your past proposals as examples for anyone else.
AI can be wrong. That's why drafts are tied to your own facts, every claim is checked, and a person on your team reviews what goes out. Converting documents to text, search and speech-to-text run on AI models hosted by Cloudflare, our hosting provider.
Accuracy first
Built to never make things up
A grant application is a promise to a funder. Grant Autopilot only states facts your own records support, and it shows you exactly what has been checked and what hasn't.
Every fact has a source
Each factual sentence links to the fact it relies on and the exact quote from your document or interview, so anyone can check it in seconds.
Only facts you've confirmed
Facts pulled from your documents are suggestions until someone on your team confirms them. A draft that leans on an unconfirmed fact is flagged.
A separate checker reads every claim
After drafting, an independent check compares each claim with its sources and makes sure every number, date and dollar amount actually appears there.
Gaps become questions, not guesses
When your records don't answer a funder's question, the draft leaves a gap and asks your team instead of filling it in.
The right amount of human review for each application
Not every application needs the same attention. Each one is routed to a review level using thresholds your organization controls.
Full review
Large requests, ones you're likely to win, or anything with unresolved flags
Your team goes through the application section by section, with every flag and source in view.
One-click approval
Most applications
A clear summary with any open questions highlighted. Approve when you're satisfied, or open the full draft.
Autopilot (optional)
Small, low-risk requests that pass every check
Only if an owner or admin turns it on and sets its limits, such as a maximum amount and a monthly cap. Anything flagged, unconfirmed or high-value still comes to you.
You approve before anything is sent.
That's the default for every organization. Autopilot stays off until an owner or admin turns it on, and you can turn it off again at any time.
Protection
How your data is protected
Grant Autopilot is built for many organizations at once, so keeping each one's data separate is part of the foundation, not an add-on.
Each organization kept separate
Every record, file and search entry belongs to exactly one organization, and every request is checked against your membership. Your files and search index are stored separately from everyone else's.
Encrypted in transit and at rest
Connections use HTTPS, and stored data is encrypted by our hosting provider. Passwords are kept only as secure hashes.
Roles for your team
Owners and admins decide who can upload, edit, review, approve or only view, and control settings, billing and deletion.
Limited staff access
Our staff look at your data only when you ask us for support, or when needed to keep the service secure or comply with the law, and only as much as needed.
Nothing sensitive in logs
Document contents are kept out of application logs and background job messages. We store a one-way hash of IP addresses, never the address itself.
Never sold, never advertised
We never sell your data or use it for advertising.
No system is perfectly secure. If a security incident affects your data, we'll notify you as the law requires.
Your data
Export and real deletion
You can take your data with you or remove it completely, at any time.
Export everything
Owners and admins can download a complete export at any time: your data as JSON plus your original files. Export files are deleted from our side 7 days after they're created.
Deletion means deletion
Delete a document and its extracted text, search entries and facts go with it. Delete your organization and we remove its files, search entries and all of its database records, including the audit log.
We keep only a minimal record that a deletion happened, plus billing records the law or our payment provider requires. Provider backups and logs roll off on their normal schedules.
You can also set a retention period for original files. After processing and that period, the originals are deleted and only the extracted text is kept.
Accountability
An audit log, and consent for every optional feature
Audit log
Consequential actions, such as uploads, approvals, submissions, exports and deletions, are written to an append-only audit log that your admins can review.
Every submission also keeps the exact package that was sent, with a fingerprint showing it hasn't changed since.
Off until you turn it on
Crawling your own website, keeping voice recordings (we otherwise keep only the transcript) and autopilot submission all stay off until an owner or admin turns them on. Turning one off stops future use.
Each choice is recorded with the policy version, the time and a hashed IP address.
Subprocessors
Who processes data for us
We share data only with these service providers, and only so they can run the service for you. We'll update this list before adding a new one.
| Subprocessor | Purpose |
|---|---|
| Cloudflare, Inc. | Hosting, database, file storage, search index, and AI models for document conversion, search embeddings and speech-to-text |
| Anthropic, PBC | The Claude language model for drafting, checking, interviews and fact extraction, through Anthropic's commercial API, which does not use the data it receives to train its models |
| Email delivery provider | Account and notification emails, and application emails when you choose email submission |
| Merchant-of-record payment processor | Checkout, payments, invoicing and sales tax |
In writing
Read the policies
Our policies are written in plain language. These are the documents that count.